Our reference: YNIMG 7778
P-authorquery-v8
AUTHOR QUERY FORM
Journal: YNIMG
Please e-mail or fax your responses and any corrections to:
E-mail: [email protected]
Fax: +1 619 699 6721
Article Number: 7778
Dear Author,
Any queries or remarks that have arisen during the processing of your manuscript are listed below and highlighted by flags in
the proof. Please check your proof carefully and mark all corrections at the appropriate place in the proof (e.g., by using onscreen annotation in the PDF file) or compile them in a separate list.
For correction or revision of any artwork, please consult http://www.elsevier.com/artworkinstructions.
Any queries or remarks that have arisen during the processing of your manuscript are listed below and highlighted by flags in
the proof. Click on the ‘Q’ link to go to the location in the proof.
Location
in article
Query / Remark: click on the Q link to go
Please insert your reply or correction at the corresponding line in the proof
Q1
Research Highlights should consist of only 85 characters per bullet point. However, the research highlights
provided for this item exceed the maximum requirement. Kindly provide the necessary corrections. For
more information, please see Guide for Authors.
Q2
Please provide the volume number and page range for the bibliography in Ref. [Okada and Delpy, 2003].
Thank you for your assistance.
Page 1 of 1
YNIMG-07778; No. of pages: 1; 4C:
NeuroImage xxx (2010) xxx
Contents lists available at ScienceDirect
NeuroImage
j o u r n a l h o m e p a g e : w w w. e l s e v i e r. c o m / l o c a t e / y n i m g
Research Highlights
1
2
5
A quantitative comparison of NIRS and fMRI across multiple cognitive tasks
6
7
Xu Cui a,b,⁎, Signe Bray a,b, Daniel M. Bryant a,b, Gary H. Glover c, Allan L. Reiss a,b
8
9
a
10
b
11
c
NeuroImage xxx (2010) xxx – xxx
Department of Psychiatry and Behavioral Sciences, School of Medicine, Stanford University, Stanford, CA 94305, USA
Center for Interdisciplinary Brain Sciences Research, Stanford University, Stanford, CA 94305, USA
Department of Radiology, Center for Advanced MR Technology, Stanford University, Stanford, CA 94305, USA
► We performed simultaneous NIRS/fMRI recording while 13 participants performed motor and cognitive tasks. ► We found that NIRS signals have weaker
signal-to-noise ratio, but are nonetheless often highly correlated with fMRI measurements. ► Both signal-to-noise ratio and the distance between the scalp
and the brain contributed to variability in the NIRS/fMRI correlations. ► In the spatial domain, we found that a photon path forming an ellipse between the
NIRS emitter and detector, with a depth of ~ 14 mm, correlated most strongly with the BOLD response.
Q1
417
18
1053-8119/$ – see front matter © 2010 Published by Elsevier Inc.
doi:10.1016/j.neuroimage.2010.10.069
Please cite this article as: Cui, X., et al., A quantitative comparison of NIRS and fMRI across multiple cognitive tasks, NeuroImage (2010),
doi:10.1016/j.neuroimage.2010.10.069
12
13
14
15
16
Supplementary materials
YNIMG-07778; No. of pages: 14; 4C:
NeuroImage xxx (2010) xxx–xxx
Contents lists available at ScienceDirect
NeuroImage
j o u r n a l h o m e p a g e : w w w. e l s e v i e r. c o m / l o c a t e / y n i m g
1
A quantitative comparison of NIRS and fMRI across multiple cognitive tasks
2
Xu Cui a,b,⁎,1, Signe Bray a,b,1, Daniel M. Bryant a,b, Gary H. Glover c, Allan L. Reiss a,b
3
4
5
a
b
c
Department of Psychiatry and Behavioral Sciences, School of Medicine, Stanford University, Stanford, CA 94305, USA
Center for Interdisciplinary Brain Sciences Research, Stanford University, Stanford, CA 94305, USA
Department of Radiology, Center for Advanced MR Technology, Stanford University, Stanford, CA 94305, USA
6
a r t i c l e
7
8
9
10
11
12
14
13
15
16
i n f o
Article history:
Received 16 August 2010
Revised 21 October 2010
Accepted 23 October 2010
Available online xxxx
a b s t r a c t
Near infrared spectroscopy (NIRS) is an increasingly popular technology for studying brain function. NIRS
presents several advantages relative to functional magnetic resonance imaging (fMRI), such as measurement
of concentration changes in both oxygenated and deoxygenated hemoglobin, finer temporal resolution, and
ease of administration, as well as disadvantages, most prominently inferior spatial resolution and decreased
signal-to-noise ratio (SNR). While fMRI has become the gold standard for in vivo imaging of the human brain,
in practice NIRS is a more convenient and less expensive technology than fMRI. It is therefore of interest to
many researchers how NIRS compares to fMRI in studies of brain function. In the present study we scanned
participants with simultaneous NIRS and fMRI on a battery of cognitive tasks, placing NIRS probes over both
frontal and parietal brain regions. We performed detailed comparisons of the signals in both temporal and
spatial domains. We found that NIRS signals have significantly weaker SNR, but are nonetheless often highly
correlated with fMRI measurements. Both SNR and the distance between the scalp and the brain contributed
to variability in the NIRS/fMRI correlations. In the spatial domain, we found that a photon path forming an
ellipse between the NIRS emitter and detector correlated most strongly with the BOLD response. Taken
together these findings suggest that, while NIRS can be an appropriate substitute for fMRI for studying brain
activity related to cognitive tasks, care should be taken when designing studies with NIRS to ensure that: 1)
the spatial resolution is adequate for answering the question of interest and 2) the design accounts for weaker
SNR, especially in brain regions more distal from the scalp.
© 2010 Published by Elsevier Inc.
37
36
35
38
Introduction
39
Cognitive tasks engender local changes in blood flow, volume, and
oxygenation in the brain. Functional magnetic resonance imaging
(fMRI) and near infrared spectroscopy (NIRS) take advantage of these
biophysical phenomena by measuring these hemodynamic correlates
of neural activity. fMRI measures the blood oxygen level-dependent
(BOLD) response that results from local concentration changes in
paramagnetic deoxy-hemoglobin (deoxy-Hb) (Ogawa et al., 1990)
and has rapidly become the gold standard for in vivo imaging of
human brain activity, due in large part to the relatively high spatial
resolution afforded by this technique. However, fMRI also presents
several challenges such as high sensitivity to participant motion, a
loud, restrictive environment, low temporal resolution, and relatively
high cost. Some of these challenges are overcome with optical
imaging: NIRS measures changes in oxygenated and deoxygenated
40
41
42
43
44
45
46
47
48
49
50
51
52
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
⁎ Corresponding author. Department of Psychiatry and Behavioral Sciences, School of
Medicine, Stanford University, Stanford, CA 94305, USA.
E-mail addresses: [email protected], [email protected] (X. Cui).
1
These authors contribute equally to this work.
hemoglobin (oxy- and deoxy-Hb) from the cortical surface and is less
invasive and expensive than fMRI.
NIRS of the human brain is a relatively flexible technology and to
date has been successfully applied in several domains. NIRS measures
changes in both oxy- and deoxy-Hb and has been used to provide
insight into the physiological mechanisms of the BOLD response
(Toronov et al., 2003; Hoge et al., 2005; Kleinschmidt et al., 1996;
Schroeter et al., 2006; Emir et al., 2008; Malonek et al., 1997; Huppert
et al., 2006b, 2007, 2009). NIRS can be very portable (Atsumori et al.,
2009) and has been proposed as a useful technology for non-invasive
brain–computer interfaces (Power et al., 2010; Coyle et al., 2004,
2007; Sitaram et al., 2007; Utsugi et al., 2008). NIRS has been used to
study brain activity in both active tasks and resting states (Boecker
et al., 2007; Herrmann et al., 2005; White et al., 2009; Honda et al.,
2010; Zhang et al., 2010; Lu et al., 2010) and may be particularly
useful with experimental paradigms that are not well suited to the
MRI scanner, such as face-to-face communication (Suda et al., 2010)
or driving (Tomioka et al., 2009).
While NIRS presents several advantages, there are also several
disadvantages to this technology. Compared to fMRI, the signal-tonoise ratio (SNR) is low and the spatial resolution of NIRS is poor. The
trajectory of the photon path from emitter to detector is assumed to
1053-8119/$ – see front matter © 2010 Published by Elsevier Inc.
doi:10.1016/j.neuroimage.2010.10.069
Please cite this article as: Cui, X., et al., A quantitative comparison of NIRS and fMRI across multiple cognitive tasks, NeuroImage (2010),
doi:10.1016/j.neuroimage.2010.10.069
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
2
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
t1:1
t1:2
t1:3
t1:4
t1:5
t1:6
t1:7
t1:8
X. Cui et al. / NeuroImage xxx (2010) xxx–xxx
be a ‘banana’ shape between the two probes (van der Zee et al., 1990;
Okada et al., 1997), but positioning the probes above target brain
regions of interest can be challenging (Kleinschmidt et al., 1996). It is
also unclear how deeply into the brain NIRS measures in practice.
NIRS signals can be conceptually compared to the BOLD response
measured with fMRI using the balloon model (Buxton et al., 1998;
Obata et al., 2004), which relates changes in BOLD to changes
in deoxy-Hb and regional cerebral blood volume (rCBV). In practice,
several studies have used simultaneous measurement of NIRS
and fMRI to better understand the generation of the BOLD response. See Steinbrink et al. (2006) for a review of combined fMRI/
NIRS studies. In one of the earliest simultaneous experiments,
Kleinschmidt et al. (1996) recorded both fMRI and NIRS signals over
ipsi- and contralateral motor cortex during a finger tapping task
and confirmed that an increase in BOLD in motor cortex contralateral to the moving hand was correlated with a decrease in deoxyHb concentration. Strangman et al. (2002) performed a similar
motor paradigm and found that oxy-Hb correlated most robustly
with the BOLD response. This may be partly attributable to a higher
SNR in oxy-Hb. Several other studies have found good correlation
between NIRS and fMRI signals (Mehagnoul-Schipper et al., 2002;
Kennan et al., 2002; Okamoto et al., 2004; Hoge et al., 2005; Toronov
et al., 2001a,b; Huppert et al., 2006b) in both simultaneous
and sequential acquisitions on a range of tasks including motor,
language, and visual tasks.
It is of great practical interest for cognitive neuroscience
researchers considering NIRS as an experimental tool to gain a better
understanding of how NIRS compares with fMRI in terms of SNR,
spatial resolution, and temporal correspondence. The majority of
combined fMRI/NIRS studies to date have been performed using
motor (Okamoto et al., 2004; Kleinschmidt et al., 1996; MehagnoulSchipper et al., 2002; Toronov et al., 2001a, 2003; Strangman et al.,
2002; Boas et al., 2003; Hoge et al., 2005; Huppert et al., 2006a,b,
2008) or visual (Toronov et al., 2007; Zhang et al., 2005a, 2006;
Abdelnour et al., 2009) tasks, and it remains unclear whether one
would draw similar conclusions using cognitive tasks with more
subtle effects. In this study we address this question by scanning
participants simultaneously with NIRS and fMRI on a battery of
cognitive tasks. We analyze the data from two perspectives, first
investigating the temporal correlation between NIRS and fMRI
signals in proximal regions of interest (ROIs) and second, exploring
the spatial distribution of fMRI voxels that best correlate with the
NIRS signal.
The data presented here are uniquely comprehensive among
combined NIRS/fMRI studies to date in that we have included
measurements from multiple brain regions, across multiple cognitive
domains, and in a relatively large number of participants. The breadth
of this data set allows us to make detailed comparisons of NIRS and
BOLD signals in terms of SNR and temporal and spatial correspondence. We hypothesized that we would find strong correlations
between the NIRS signals and proximal fMRI voxels, but that this
correlation might be stronger in tasks with higher SNR, and weaker
when the photons travel a larger distance from the scalp to the brain's
surface. In the spatial domain, we explored several possible ROI
shapes, to empirically confirm the effective measurement area of NIRS 129
channels.
130
Methods
131
Participants
132
Thirteen healthy young adults (mean age 27.9, age range 21–42, 6
males) participated in this study using simultaneous fMRI and NIRS.
Written informed consent was obtained from all participants, and the
study protocol was approved by the Stanford University Institutional
Review Board.
133
Experimental procedure
138
Participants performed four experiments while being scanned by
fMRI and NIRS simultaneously: left finger tapping (tap), go/no-go
(nog), judgment of line orientation (jlo), and an N-back working
memory task using visuospatial stimuli (vis); these tasks are described
in detail below and have been used in previous studies from our group
(Hoeft et al., 2007; Kesler et al., 2004; Haberecht et al., 2001).
Participants were given task instructions verbally before entering the
scanner. Eleven participants completed 4 out of 4 tasks, one completed
3, and one completed 2 (Table 1). Two participants did not complete all
4 tasks due to discomfort induced by the NIRS probes. The order of
experiments was fixed: tap, nog, jlo, and vis. All experimental stimuli
were presented using E-prime software (http://www.pstnet.com/).
139
140
Task descriptions
151
Finger tapping (tap)
The finger tapping task consisted of 10 alternating tapping and
resting epochs. Each tapping epoch lasted 15 s and each resting epoch
lasted 20 s. During the tapping epochs, a circular checkerboard
pattern flashed on the screen. During the resting epochs, a fixation
cross was displayed on the screen. Participants were instructed to tap
the fingers on their left hand vigorously while the checkerboard was
flashing and remain still while the fixation cross was displayed.
152
Go/no-go task (nog)
The go/no-go task consisted of rest, go (G), and no-go (NG) epochs
in the following order: Rest–G–NG–G–NG–G–NG–G–NG–G–NG–G–
NG–Rest. Each rest epoch lasted 30 s, during which subjects passively
viewed a blank screen; each task epoch lasted 26 s. During the go and
no-go epochs, on each trial participants were presented with 1 of 12
different letters for 500 ms, followed by a 1.5 s interstimulus interval.
During go epochs, participants were instructed to press a button with
their right index finger each time a letter appeared on the screen; at
the start of these epochs the instruction, “Push for all letters” was
displayed for 2 s. During no-go epochs, participants were instructed to
push the button for all letters except X, for which they should
withhold their response; these epochs began with a 2 s display of the
instructions, “Push for All Letters Except X.” During no-go epochs, the
letter X was presented on 50% of trials.
160
161
Table 1
Experimental tasks and number of participants.
Task
Description
Number of participants
Number of NIRS channels per participant
Total number of NIRS time series
(# participants × # channels)
tap
nog
jlo
vis
Tapping of fingers on left hand
Go/no-go
Judgment of line orientation
Visuospatial N-Back working memory task
13
13
12
11
24
24
24
24
Total
312
312
288
264
1176
Please cite this article as: Cui, X., et al., A quantitative comparison of NIRS and fMRI across multiple cognitive tasks, NeuroImage (2010),
doi:10.1016/j.neuroimage.2010.10.069
134
135
136
137
141
142
143
144
145
146
147
148
149
150
153
154
155
156
157
158
159
162
163
164
165
166
167
168
169
170
171
172
173
174
X. Cui et al. / NeuroImage xxx (2010) xxx–xxx
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
Judgment of line orientation task (jlo)
The judgment of line orientation task consisted of rest, experimental (E), and control (C) epochs in the following order: Rest–E–C–
E–C–E–C– Rest–E–C–E–C–E–C –Rest. Each rest epoch lasted 30 s,
during which subjects passively viewed a blank screen. Control and
experimental epochs consisted of 10 stimuli presented for 2.5 s each,
with a 300 ms interstimulus interval. On all trials two sets of lines
appeared on the screen: the top set of purple lines were arranged in
the shape of a protractor, and the lower set of lines were either a
subset of the angled lines presented above or, during control epochs,
simply three parallel lines. Control epochs began with a 4 s display of
the instructions, “Judge if the colors match.” During these epochs,
participants were instructed to press a button if the color of the lines
on the lower part of the screen matched the color of the lines on the
upper part of the screen. Experimental epochs began with a 4 s display
of the instructions, “Judge if the line orientations are equal.” Here the
task was to respond with a button press if the lines presented on the
bottom of the screen were in the same orientation as the two lines
highlighted in yellow in the protractor on the upper portion of the
screen. There were two levels of difficulty for the experimental
epochs: the first three experimental epochs were less difficult, and the
final three were more difficult. For the easy experimental task, the
protractor on the upper portion of the screen was made up of 5 lines,
while for the difficult task it was made up of 11 lines. In addition, in
the difficult task the length of the target lines was shorter.
Visuospatial N-back working memory task (vis)
During the visuospatial working memory tasks, 1-back (1B), 2back (2B), and control (C) epochs were presented in the following
order: rest–1B–C–1B–C–1B–C–rest–2B–C–2B–C–2B–C–rest. Each rest
3
epoch was 16 s long, during which subjects passively viewed a blank
screen. Experimental epochs began with a 2 s display of the
instruction: “Push for 1 Back” in the 1-back task, and “Push for 2
Back” in the 2-back task. Control epochs began with a 2 s display of
the instruction, “Push for Center.” Each control and experimental
epoch consisted of 14 stimuli presented for 1.5 s each, with a 1 s
interstimulus interval. The stimulus was the letter ‘O’ presented in
one of nine locations in a 3 × 3 matrix. In the 1-back task, participants
were instructed to respond if the stimulus was in the same location
as in the previous trial. In the 2-back task, participants were instructed
to respond if the stimulus was in the same location as in two
trials previously. In the control task, participants were instructed to
respond if the stimulus appeared in the center of the screen.
204
205
NIRS data acquisition
217
We used an ETG-4000 (Hitachi Medical, Japan) Optical Topography
system to measure the concentration change of oxy-Hb and deoxy-Hb
at a rate of 10 Hz. This device uses the Modified Beer-Lambert law
(MBLL) (Cope and Delpy, 1988; Cope, 1991) to calculate changes in
chromophore concentrations from the attenuation of light entering
the head at multiple wavelengths. Two “3 × 3” measurement patches
provided by Hitachi were attached to a regular swimming cap
(Fig. 1A), and elastic straps were used to ensure good contact of the
measurement probes on the participants’ heads. Five emitters and four
detectors were positioned alternatingly in each patch, for a total of 18
probes, resulting in 24 measurement channels (Figs. 1A, D, and E). The
distance between each emitter–detector pair
pffiffiffiwas 3 cm. The distance
between adjacent channels was thus 1:5 × 2 = 2:1 cm (e.g., Fig. 1E
between channel 1 and 3) or 3 cm (e.g., between channel 1 and 2). MRI
218
Fig. 1. NIRS measurement configuration. (A) We used two sets of 3 × 3 measurement patches, one placed over the right frontal cortex, the other over the right parietal cortex. A
swimming cap was used to hold the probes against the scalp. Probes (not shown) were plugged in the red or blue holes on the patches. (B) MRI fiducial markers (in white) were
affixed to the inner surface of the measurement patch at the locations of the 24 channels (between every emitter–detector pair). The ring-shaped markers allowed us to precisely
identify the channel positions. (C) MRI fiducial markers (indicated by arrows) are easily identified in the structural MRI images. (D) A rendering of the skull surface showing the 24
channel markers in one participant, obtained from the T1-weighted image volume. (E) Schematic of the probe/channel configurations on the head. Red circles indicate the photon
emitters; blue circles indicate the photon detectors; measurement channels (numbered in square rectangle) are located between emitter/detector pairs. (For interpretation of the
references to colour in this figure legend, the reader is referred to the web version of this article.)
Please cite this article as: Cui, X., et al., A quantitative comparison of NIRS and fMRI across multiple cognitive tasks, NeuroImage (2010),
doi:10.1016/j.neuroimage.2010.10.069
206
207
208
209
210
211
212
213
214
215
216
219
220
221
222
223
224
225
226
227
228
229
230
231
4
X. Cui et al. / NeuroImage xxx (2010) xxx–xxx
when calculating the noise level in NIRS signal) to remove highfrequency instrument noise and low frequency drift in oxy-Hb
and deoxy-Hb signals. NIRS time courses in each experiment, in
each channel, were then temporally aligned with the corresponding
fMRI data and resampled at the fMRI sampling rate of 0.5 Hz. TotalHb was calculated as the sum of oxy-Hb and deoxy-Hb. As deoxy-Hb
usually decreases during neural activation, the sign of the deoxy-Hb
signal was flipped for easier comparison with oxy-Hb and the BOLD
signal.
290
291
fMRI preprocessing
Functional MRI images were realigned, slice timing corrected, and
coregistered to the structural images using SPM software (version 8,
http://www.fil.ion.ucl.ac.uk/spm/). The fMRI images were not normalized to a standard template or spatially smoothed. Temporal drift
in the fMRI signal was removed by subtracting a moving average with
a window size of 50 TRs (100 s).
299
247
248
fiducial markers (MRIEquip, SKU: BP-103X, Model: BP-1015, 15-mm
outer diameter, 3.5 mm thick, 5-mm central axis hole, https://www.
mriequip.com/store/pc/viewPrd.asp?idcategory=72&idproduct=270)
were affixed to the inner side of the measurement patch at the
location of each of the 24 channels (i.e., between every emitter–
detector pair) to identify the positions of NIRS channels in the MRI
structural images. The ring-shaped markers allowed us to precisely
identify the channel positions. The measurement area covered the
right inferior frontal cortex and right parietal cortex (Figs. 1A, D,
and E). The cap remained in the same position on the participant's
head for all 4 experimental tasks.
The ETG-4000 was located in the control room of the MRI scanning
facility. A set of long optical fibers (10 m, Hitachi Medical, Japan) were
passed through a waveguide in the wall between the control room
and the MRI scanner room. At the beginning of each functional scan,
the E-prime stimulus program triggered both the ETG-4000 and fMRI
scanner simultaneously via serial port.
249
fMRI data acquisition
306
250
262
Structural and functional MRI images were acquired using a 3 T
Signa Discovery 750 (GE Medical Systems). The acquisition parameters
for axial-oblique 3D T1-weighted structural images were: fast spoiled
gradient recalled echo (FSPGR) pulse sequence, inversion recovery
preparation pulse TI= 400 ms; repetition time (TR) = 8.5 ms; echotime (TE) = 3.4 ms; flip angle = 15°; receiver bandwidth ±32 kHz;
slice thickness = 1.2 mm; 128 slice locations; number of excitations
(number of signals averaged) = 1; field-of-view (FOV)= 22 cm; phase
FOV 0.75; acquisition matrix= 256 × 192. fMRI data acquisition
parameters: T2*-sensitive gradient echo spiral pulse sequence sensitive to BOLD contrast (Glover and Law, 2001), TR = 2000 ms;
TE= 30 ms; flip angle = 80°; FOV = 220 mm × 220 mm; 31 slices. The
size of each voxel is 3.4 × 3.4 × 4 mm.
263
Identifying the channel locations
Region of interest (ROI)
We were interested in assessing the correlation between the time
series measured with NIRS and the fMRI voxels adjacent to the NIRS
recording site. To obtain this correlation, we first identified a region of
interest (ROI) for each channel in each subject. This ROI, in theory,
should be the path along which photons travel from the emitter to the
detector. Without knowing the exact shape of the path, we used a
spherical region underneath the channel as an approximation. For each
channel, we projected the channel marker from the scalp to the brain
surface and grew a spherical region around the projection point with a
radius of 5 voxels (~17.5 mm). Then we removed any portion of the
sphere that fell outside the brain mask and only kept the voxels inside
the brain. fMRI signals from the voxels in this ROI were extracted and
averaged. Thus, for each NIRS time course, we obtained a corresponding
BOLD fMRI time course. Accounting for all subjects and tasks, we
obtained 1176 such pairs in total, and we subsequently refer to each
pair as an ‘instance.’ Note that this procedure therefore included
instances with and without task-related brain activation.
264
265
fMRI–NIRS correlation
For each channel and each task in each subject, the Pearson
correlation coefficient was calculated between the NIRS time courses
(oxy-Hb, deoxy-Hb, and total-Hb) and the fMRI time course in the
corresponding ROI. We have 1176 correlation coefficients for each of
the oxy-, deoxy-, and total-Hb (Table 1).
324
325
270
We used the fiducial markers, visually evident in the MRI images,
to manually identify the center position of each of the 24 channel
markers on each subject (Fig. 1B–D) and recorded the coordinates
using the xjView toolbox (http://www.alivelearn.net/xjview). The
probe positions (emitters and detectors) were calculated from the
channel positions based on the cap configuration (Fig. 1E) using linear
interpolation or extrapolation.
271
Scalp–brain distance
330
331
272
Contrast-to-noise ratio (CNR)
We used the contrast-to-noise ratio (CNR) to quantify the signalto-noise ratio (Zhang et al., 2005b) in both the fMRI and NIRS time
courses. The CNR calculates the amplitude difference between the
signal during task blocks and control or rest blocks, divided by the
pooled standard deviation. Larger CNR indicates that the ratio of taskrelated signal to noise is larger.
335
336
285
We created a skull-stripped brain mask for each subject. For each
channel marker, we projected the marker from the scalp to the brain
surface by finding the point on the brain surface which was closest
to the marker. The distance between the channel marker and its
projected point on the brain surface is what we refer to as the ‘scalp–
brain distance.’ This distance is determined by the thickness of the
skull, but also by hair thickness and cerebrospinal fluid.
To calculate the scalp–brain distance on the standard template
(colin27, the T1.nii template included with SPM, http://www.fil.ion.
ucl.ac.uk/spm/) as shown in Fig. 8, we generated 2701 points
uniformly distributed on the scalp and projected these points on the
brain surface. For each point, we calculated the scalp–brain distance
using the method described above. We then visualized the distance
with colors on the surface of the scalp (Fig. 8).
Noise level in NIRS signal
To quantify the high-frequency noise level in the NIRS data, we
used high-pass filtering to remove the components with frequency
less than 2 Hz from the raw NIRS signal. We calculated the variance of
the remaining signal as an indicator of the noise level.
286
Temporal analysis
287
288
NIRS data processing
A band-pass filter with cutoff frequencies 0.01 and 0.5 Hz was
applied to the raw NIRS data before any further processing (except
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
251
252
253
254
255
256
257
258
259
260
261
266
267
268
269
273
274
275
276
277
278
279
280
281
282
283
284
292
293
294
295
296
297
298
300
301
302
303
304
305
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
326
327
328
329
332
333
334
337
338
339
340
341
meanðdur Þ meanðpreÞ
CNR ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
var ðdur Þ + var ðpreÞ
343
342
289
The term “dur” refers to signal during task blocks, and “pre” refers
to signal during the control or rest period before the task blocks.
Considering the hemodynamic delay, we chose a time window from 6
to 12 s after the beginning of task blocks as “dur” and 0–5 s before task
as “pre.”
Please cite this article as: Cui, X., et al., A quantitative comparison of NIRS and fMRI across multiple cognitive tasks, NeuroImage (2010),
doi:10.1016/j.neuroimage.2010.10.069
344
345
346
347
348
X. Cui et al. / NeuroImage xxx (2010) xxx–xxx
349
Spatial analysis
350
351
For the spatial analysis we explored the spatial correspondence
between NIRS and fMRI measurements. NIRS measures changes in oxyand deoxy-Hb concentrations along the photon path from emitter to
detector. It is generally believed that this path forms a ‘banana’ shape
between the emitter and detector. As we were not able to measure the
photon paths directly, we approached this question indirectly. We
assumed that if the fMRI signal in a voxel or a set of voxels is highly
correlated with the NIRS signal, this voxel or set of voxels was likely in
the photon path. Thus, we asked, “where are the voxels whose fMRI
signal is best correlated to the corresponding NIRS signal?” To answer
this question, we tried two approaches. Our first approach was to find
the single voxel which best correlated with the NIRS signal for each
channel, which we call the local best correlating voxel (LBCV), and then
observe the spatial distribution of these voxels. While the photon path
covers many more than one voxel, we reasoned that the distribution of
LBCVs across all 1176 instances would reflect the photon path pattern in
the brain. While the distribution of LBCVs gave us a general idea of where
the NIRS measurement occurs, in reality photons always travel through
multiple voxels. In the second approach, we systematically varied the
shape of the ROI and studied how this affected the correlation with NIRS
signals. While there are an infinite number of possible shapes, we chose
to focus on three shapes: a sphere, a spherical shell, and a 1-D elliptical
ring. We chose these shapes based on previous work examining the
photon scattering pattern (Okada et al., 1995; Okada and Delpy, 2003).
We systematically varied the size and orientation of the shapes,
extracted fMRI signals from the resultant ROIs, and determined the
size or orientation which gave the best fMRI–NIRS correlation, for each
channel. While these approaches could not decisively determine the
photon path, we nonetheless expected to gain some understanding of
the corresponding spatial pattern of this path.
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
5
Finding the local best correlating voxel (LBCV)
For each channel, we computed the correlation coefficient
between the NIRS signals (both oxy-Hb and deoxy-Hb) and the
fMRI signal from each voxel in the entire brain to create a correlation
image. We then smoothed the correlation image with an 8-mm
Gaussian kernel (Fig. 2A), which allowed us to be less sensitive to
noise in later steps. We identified voxels whose correlations were
(1) local maxima and (2) above 0.1 (Fig. 2B). If more than one voxel
satisfied these conditions, we chose the voxel closest to the channel's
projection point on the brain (based on the assumption that the
photon path should be close to the channel).
To quantify the spatial location of LBCVs relative to the channels and
brain surface, we calculated the distance between every two of the four
key points shown in Fig. 2C: point A, the channel marker; point B, the
projection point of the channel marker (A) on the brain surface; point C,
the LBCV; and point D, the shortest path from point C to the brain
surface. We also calculated the angle, α, between AB and BC.
380
381
Finding the optimal spherical volume
For each channel, we identified the projection point on the brain
and grew a spherical ROI around this point into the brain. We varied
the radius of this sphere from 1 to 10 voxels (note: in the temporal
analysis, the radius was fixed at 5 voxels). We then extracted the fMRI
signal from the ROI and calculated the correlation with the oxy-Hb
NIRS signal from the corresponding NIRS channel.
397
Finding the optimal spherical shell
For each channel, we identified the projection point on the brain
surface and generated an ROI in the shape of a spherical shell around
this point (thickness: 1 voxel; for a schematic, see Results: Fig. 13A).
We varied the radius of this sphere from 1 to 10 voxels, extracted the
404
Fig. 2. Identifying the spatial location of the LBCV. (A) For a given NIRS channel, we calculated the correlation between the measured signal and the BOLD signal at every voxel in the
brain. This generated a correlation image which was then smoothed with an 8-mm Gaussian kernel. (B) 1-D schematic demonstrating the identification of the LBCV above a
threshold of 0.1. The horizontal dotted line indicates the threshold (0.1). The red arrows indicate the local peaks. The shortest distance (e.g., leftmost) supra-threshold peak is the
LBCV. (C) Schematic drawing of the distances and angles to be calculated. Point A is the channel location on the scalp, as identified by the MRI fiducial marker. Projecting point A onto
the brain surface we get point B. We find the LBCV, point C, and projecting C onto the brain surface we get point D. (For interpretation of the references to colour in this figure legend,
the reader is referred to the web version of this article.)
Please cite this article as: Cui, X., et al., A quantitative comparison of NIRS and fMRI across multiple cognitive tasks, NeuroImage (2010),
doi:10.1016/j.neuroimage.2010.10.069
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
398
399
400
401
402
403
405
406
407
408
6
X. Cui et al. / NeuroImage xxx (2010) xxx–xxx
409
410
fMRI signal from the resulting ROI, and calculated the correlation with
the oxy-Hb NIRS signal from the channel.
411
412
421
422
Finding the optimal elliptical ring
For each channel, we identified the position of the corresponding
emitter and detector and projected the two points onto the brain
surface. We used these as antipodal points and constructed a series
of elliptical rings by varying two parameters: depth and angle off the
vertical plane (for a schematic, see Results: Fig. 13B). The vertical
plane is the plane defined by the emitter, the detector, and the
channel projection on the brain surface. The depth ranges from 1 to 10
voxels, and the angle ranges from −90° to 90° with a step size of 30°.
The thickness of the ring was fixed at 1 voxel. We then extracted the
fMRI signal from these ROIs and calculated the correlation with the
oxy-Hb NIRS signal from each corresponding channel.
423
General linear model (GLM)-based analysis in each modality
424
425
fMRI
Functional MRI images were realigned, slice timing corrected,
coregistered to the structural images, normalized to the standard
template and spatially smoothed (8-mm) using SPM software
(version 8, http://www.fil.ion.ucl.ac.uk/spm/).
We constructed a general linear model (GLM) in SPM for each
subject and for each task. For the tapping task, the design matrix
413
414
415
416
417
418
419
420
426
427
428
429
430
A
0.87
consisted of one regressor with onsets for the tapping blocks
(duration 15 s), while for the other tasks two regressors were
included, one for the onset of all task blocks (including all difficulty
levels) and one for control blocks; regressors were convolved with the
canonical hemodynamic response function (HRF). Following estimation, we generated a contrast image comparing the task and control
conditions (or simply task vs. implicit baseline for the tapping task).
Finally, to estimate group-level effects, we performed one-sample ttests on the contrast images across subjects.
431
432
NIRS
A band-pass filter with cutoff frequencies 0.01 and 0.5 Hz was
applied to the raw NIRS data. For each task, we generated a similar
model, as described above for the fMRI analysis, and entered this into
a linear regression for each channel (all regressors were convolved
with the canonical HRF). Contrast values were calculated based on the
parameter estimates. For the group analysis, channel positions were
normalized to a standard template using the transform matrix
generated by the normalization procedure in the fMRI data preprocessing. A contrast image was created using the contrast values
and their corresponding channel positions on the brain surface. Cubic
spline interpolation was applied to estimate the contrast value in
voxels between channels. Finally, a t-test was performed on the
contrast images across subjects.
440
441
B
0.82
0.83
0.77
C
D
0.00
0.00
-0.02
0.08
E
F
0.74
0.11
0.07
0.74
G
H
Channel
0.25
oxy-Hb
deoxy-Hb
BOLD
Skull
0.28
Brain
ROI
Fig. 3. Examples of oxy-Hb, deoxy-Hb, and BOLD fMRI time courses to demonstrate the wide spectrum of NIRS–fMRI correlations. Time courses presented here were normalized to
have a standard deviation of 1. Oxy-Hb, deoxy-Hb, and BOLD signals are plotted in red, blue, and black, respectively. As deoxy-Hb usually decreases during neural activation, the sign
of the deoxy-Hb signal was flipped for easier comparison with oxy-Hb and the BOLD signal. (A) An example of high correlation and high CNR (subject: 12, task: tap, channel: 9). (B) An
example of high correlation but low CNR (subject: 4, task: nog, channel: 21). (C) An example of high BOLD CNR but low NIRS CNR (subject: 6, task: tap, channel: 4). (D) An
example of low BOLD CNR but high NIRS CNR (subject: 7, task: tap, channel: 13). (E) An example of high BOLD-oxy-Hb correlation but low BOLD-deoxy-Hb correlation
(subject: 13, task: vis, channel: 24). (F) An example of low BOLD-oxy-Hb correlation but high BOLD-deoxy-Hb correlation (subject: 8, task: vis, channel: 12). (G) An example of
low BOLD-oxy-Hb correlation and low BOLD-deoxy-Hb correlation (subject: 6, task: tap, channel: 22). (H) Schematic of the ROI from which the fMRI signal was extracted.
Correlation coefficients appear in the top right corner of each panel; vertical dotted lines indicate the onset timing of task or control blocks. (For interpretation of the references to
colour in this figure legend, the reader is referred to the web version of this article.)
Please cite this article as: Cui, X., et al., A quantitative comparison of NIRS and fMRI across multiple cognitive tasks, NeuroImage (2010),
doi:10.1016/j.neuroimage.2010.10.069
433
434
435
436
437
438
439
442
443
444
445
446
447
448
449
450
451
452
453
X. Cui et al. / NeuroImage xxx (2010) xxx–xxx
A
7
C
oxy-Hb
0.2
25%
BOLD-NIRS
correlation
(all)
-0.8 -0.4
0.1
D
0
0.4
0.8
deoxy-Hb
20%
0
B
-0.8 -0.4
0.6
E
BOLD-NIRS
correlation
(top 100)
0
0.4
0.8
total-Hb
15%
0.4
0.2
-0.8 -0.4
0
0.4
0.8
BOLD-NIRS correlation
oxy-Hb
deoxy-Hb
total-Hb
Fig. 4. Mean and distribution of fMRI–NIRS correlations. (A) The mean and standard error of correlations between BOLD and oxy-Hb, deoxy-Hb, and total-Hb are 0.26± 0.008, 0.23±
0.008, and 0.19 ± 0.008, respectively. The BOLD-oxy-Hb correlation is significantly higher than BOLD-deoxy-Hb correlation (p = 2.6 × 10−4, paired t-test, df= 1175) and the BOLD-totalHb correlation (p b 10−10). (B) The mean and standard error of the top 100 values are 0.74 ± 0.006, 0.74± 0.005, and 0.66± 0.006 for oxy-, deoxy-, and total-Hb, respectively. (C–E) The
distribution of BOLD–NIRS correlations. The percentage of high correlation (higher than 0.5) instances are 25%, 20%, and 15% for oxy-Hb, deoxy-Hb, and total-Hb, respectively. Vertical
dotted lines indicate 0 NIRS–fMRI correlation. Oxy-Hb and deoxy-Hb yield similar correlations with the BOLD signal, while total-Hb shows lower correlations.
454
Results
455
Temporal analysis
456
fMRI–NIRS correlations show wide variability
We found a wide range of correlations between signals in our 1176
NIRS–fMRI pairs. To demonstrate this variability we plotted a selection
of time courses showing strong and weak correlations (Fig. 3). In some
panels the NIRS–fMRI correlations are as high as 0.8 (Fig. 3A and B),
but in others as low as 0 (Fig. 3C and D). In some cases oxy-Hb
correlates with the BOLD signal better than deoxy-Hb (Fig. 3E) and in
some other cases, deoxy-Hb and BOLD correlations are higher than
those observed with oxy-Hb (Fig. 3F). High correlations between BOLD
and NIRS sometimes occur in brain areas which respond to the
experimental task (high CNR, Fig. 3A), but also sometimes in regions
which do not correlate well with the task (low CNR, Fig. 3B).
Though NIRS–fMRI correlations show wide variability in magnitude, the mean correlation is significantly higher than 0 (Fig. 4A). The
average correlation between BOLD and oxy-Hb is 0.26, and between
BOLD and deoxy-Hb is 0.23, across all 1176 instances. The BOLD-oxy-
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
Cognitive task does not affect NIRS/BOLD correlations
We have shown that NIRS–fMRI correlations are significantly
greater than 0, but vary widely. Thus, we asked whether there was a
difference in these correlations between the motor task (tap) and the
483
oxy-Hb
deoxy-Hb
total-Hb
0.3
BOLD-NIRS correlation
457
458
Hb correlation is significantly higher than BOLD-deoxy-Hb correlation
(p = 2.6 × 10−4, paired t-test, df = 1175) and the BOLD-total-Hb
correlation (p b 10−10). Total-Hb shows poor correlation with BOLD,
compared to oxy-Hb and deoxy-Hb. Due to low SNR, correlations may
be low in task-irrelevant channels. Therefore we also plotted the
average correlation in the channels that show the highest correlations.
Fig. 4B shows the mean correlation from the top 100 of the 1176
values (8.5%). Among the instances showing the strongest correlations, both oxy-Hb and deoxy-Hb show a correlation higher than 0.7
with the fMRI signal. The distributions of correlation coefficients are
plotted in Fig. 4C–E for oxy-Hb, deoxy-Hb, and total-Hb, respectively.
0.2
0.1
0
tap
nog
jlo
vis
Fig. 5. Mean and standard error of the BOLD–NIRS correlations for all four tasks. There
was no significant difference in the BOLD–NIRS correlations between the tasks. tap,
tapping of fingers on left hand; nog, go/no-go; jlo, judgment of line orientation; vis,
visuospatial N-back working memory task.
Fig. 6. fMRI-oxy-Hb correlation for each channel in different tasks. In all four tasks, the
correlation is higher in the middle frontal area and the inferior parietal area than other
areas. The location of each channel for individual participants was mapped to a
standard brain, and then a single position for each channel was calculated by averaging
across the 13 participants. tap, tapping of fingers on left hand; nog, go/no-go; jlo,
judgment of line orientation; vis, visuospatial N-back working memory task.
Please cite this article as: Cui, X., et al., A quantitative comparison of NIRS and fMRI across multiple cognitive tasks, NeuroImage (2010),
doi:10.1016/j.neuroimage.2010.10.069
474
475
476
477
478
479
480
481
482
484
485
486
8
X. Cui et al. / NeuroImage xxx (2010) xxx–xxx
Fig. 7. CNR in fMRI, oxy-Hb, and deoxy-Hb for each channel in different tasks. CNR values are normalized within each panel so the minimum is 0 and the maximum is 1. Channel
significance values are plotted in Fig. S1. tap, tapping of fingers on left hand; nog, go/no-go; jlo, judgment of line orientation; vis, visuospatial N-back working memory task.
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
cognitive tasks (nog, jlo and vis). To address this question, we plotted
the BOLD–NIRS correlations for the four experimental tasks (Fig. 5).
Results indicated that there were no significant differences between
any of the tasks (two sample t-tests were performed between every
pair of tasks for each hemoglobin species; p N 0.1 for all comparisons,
except the nog and vis comparison in fMRI-total-Hb, where p = 0.02).
We further found that the spatial pattern of NIRS–fMRI correlations is similar across different experimental tasks. Fig. 6 shows the
correlation between fMRI and oxy-Hb on a template brain, in each
channel, for each of the 4 tasks. The middle frontal area and the
inferior parietal area show a higher correlation than other areas, and
this is generally true for all tasks.
CNR, on the other hand, is highly dependent on the experimental
task. Fig. 7 shows the color-encoded CNR on a template brain for each
task. As expected, the CNR pattern is different for different tasks. In
the tap task, the motor area has higher CNR as measured by both fMRI
and NIRS, with the fMRI pattern more focal. For the cognitive tasks,
fMRI and NIRS CNR maps are less similar; in Fig. S1, we show 504
significance levels for the fMRI and NIRS channels.
505
Longer scalp–brain distance degrades NIRS–fMRI correlations
We investigated factors that may contribute to the observed
variability in NIRS–fMRI correlations. As NIRS measurements critically
depend on the infrared photons passing through the skull and brain
tissues, it is likely that the scalp–brain distance, which affects the
photon path, also affects the NIRS signal quality and thus the correlation between NIRS and fMRI. To assess this possibility, we calculated the scalp–brain distance for every channel and correlated this
distance with a measure of NIRS data quality and the strength of NIRS–
fMRI correlations.
The scalp–brain distance varies across brain regions. In Fig. 8 we
show the scalp–brain distances for a template brain (colin27, the T1.
nii template included with SPM, http://www.fil.ion.ucl.ac.uk/spm/). It
can be seen that parietal regions have the longest scalp–brain distance
Fig. 8. Scalp–brain distance of a standard brain (colin27). The mean scalp–brain distance is 16.8 mm. The scalp–brain distance is smaller in frontal regions and larger in parietal regions.
Please cite this article as: Cui, X., et al., A quantitative comparison of NIRS and fMRI across multiple cognitive tasks, NeuroImage (2010),
doi:10.1016/j.neuroimage.2010.10.069
506
507
508
509
510
511
512
513
514
515
516
517
518
519
X. Cui et al. / NeuroImage xxx (2010) xxx–xxx
B
0.6
1
BOLD-NIRS correlation
Variance of NIRS oxy-Hb noise
A
9
0.1
0.01
0.001
0.4
0.2
-10
0.0 oxy-Hb p < 10
deoxy-Hb p < 10 -10
total-Hb
p = 10 -6
-0.2
10
14
18
22
26
10
Scalp-brain distance (mm)
14
18
22
26
Scalp-brain distance (mm)
Fig. 9. High-frequency noise in NIRS and BOLD–NIRS correlation as functions of scalp–brain distance. (A) High-frequency noise in NIRS signal and scalp–brain distance are correlated.
The scalp–brain distances are binned in 1-mm increments. Circle area is proportional to the sample size. Error bars indicate the standard error. The y-axis is on a logarithmic scale. (B)
Scalp–brain distance affects BOLD–NIRS correlation. The BOLD–NIRS correlation is more likely to be low in regions where the scalp–brain distance is higher.
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
(N18 mm), while frontal and temporal regions have the shortest
(b16 mm). This difference was confirmed across our subject groups,
where longer scalp–brain distances in parietal channels were
observed relative to frontal channels (Fig. S8).
We found that the scalp–brain distance has a direct impact on NIRS
data quality as indicated by the noise level in the NIRS data (Fig. 9A).
For channels where the scalp–brain distance is larger than 22 mm, it is
likely that the measured NIRS signals contain increasing highfrequency noise.
Next, we found that the scalp–brain distance affects the NIRS–fMRI
correlation. In Fig. 9B we plot the BOLD–NIRS correlation as a function
of the scalp–brain distance. For all of the measured hemoglobin
signals (oxy-, deoxy-, and total-Hb), the NIRS–fMRI correlation is
significantly negatively correlated with the scalp–brain distance
(p b 10−6). The variability of the NIRS–fMRI correlations (Figs. 3 and
4) can thus be partly explained by the scalp–brain distance.
A
CNR
(all)
Higher CNR yields higher NIRS–fMRI correlations
We have shown that long scalp–brain distances degrade NIRS
data quality and reduce the correlation between NIRS and fMRI.
Another factor that could affect the NIRS–fMRI correlation is the true
signal level in the data. While fMRI and NIRS both measure
the hemodynamic response, they are unlikely to be correlated if
the underlying hemodynamic signal does not change. For example, if
a brain region does not respond to the experimental task, it is likely
that the measured data in this region is primarily noise, and thus
the NIRS–fMRI correlation may be low. To test this hypothesis, we
first calculated the CNR value for each of the 1176 pairs of time
courses and next tested how the CNR correlates with the NIRS–fMRI
correlations.
We calculated the CNR of the fMRI and NIRS signals and found that
the fMRI CNR is significantly higher than the CNRs of oxy-Hb and
deoxy-Hb (p b 10−10, paired t-test, df= 1175). In Fig. 10 we plot the
C
0.3
BOLD
0.2
-2
0.1
-1
0
1
2
D
3
4
oxy-Hb
B
1.6
-2
CNR
(top 100)
1.2
-1
0
1
E
2
3
4
deoxy-Hb
0.8
0.4
0
BOLD
oxy-Hb
deoxy-Hb
-2
-1
0
1
2
3
4
CNR
Fig. 10. Mean and distribution of the CNR in BOLD, oxy-Hb, and deoxy-Hb. (A) Mean CNR of all 1176 pairs of time courses. BOLD signal on average has the highest CNR (0.36 ± 0.02,
standard error). The CNR of oxy- and deoxy-Hb are 0.12 ± 0.01 and 0.13 ± 0.01, respectively. The BOLD CNR is significantly higher than that of oxy-Hb (p b 10−10, paired t-test,
df = 1,175) and that of deoxy-Hb (p b 10−10). There is no significant difference between the CNRs of oxy-Hb and deoxy-Hb (p = 0.58). (B) The mean and standard error of the top 100
values are 1.75 ± 0.05, 1.11 ± 0.026, and 1.1 ± 0.027 for BOLD, oxy-Hb, and deoxy-Hb, respectively. (C–E) The distribution of CNR in BOLD, oxy-Hb, and deoxy-Hb, respectively.
Vertical dotted lines indicate CNR = 0.
Please cite this article as: Cui, X., et al., A quantitative comparison of NIRS and fMRI across multiple cognitive tasks, NeuroImage (2010),
doi:10.1016/j.neuroimage.2010.10.069
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
10
554
555
556
557
558
559
560
mean and distribution of the CNR values. In Fig. 10A we show the
mean CNR for all 1176 instances. We also show that the mean of
the 100 largest fMRI CNR values is also significantly higher than the
mean of the 100 largest oxy-Hb or deoxy-Hb CNR values. This result
indicates that the CNR of fMRI is generally better than that of NIRS.
We expected that NIRS–fMRI correlations would be higher for time
courses with higher CNR. To test this hypothesis, we plotted the BOLD–
NIRS correlations as a function of CNR (Fig. 11). We found that the
BOLD–NIRS correlation is significantly positively correlated with the
oxy-Hb
A
deoxy-Hb
0.4
total-Hb
p = 8x10-6
0.3
p = 0.06
0.2
p = 2x10-5
0.1
-1
-0.5
0
0.5
1
B
0.5
0.4
p = 2x10-10
0.3
p = 5x10-7
0.2
p = 3x10-6
0.1
-1
-0.5
0
0.5
1
1.5
oxy-Hb CNR
C
CNR. It is notable that the BOLD–NIRS correlation shows some nonlinearity as the BOLD CNR becomes increasingly negative (Fig. 11A).
The variability of the NIRS–fMRI correlation is thus partially explained
by the signal level. We also found that, for instances with high BOLD
CNR, NIRS CNR is likely to be higher (p b 10−10, Fig. S2).
561
562
Spatial analysis
566
Spatial properties of local best correlating voxels (LBCVs)
In Fig. 12 we show the position of LBCVs relative to the emitter and
detector. In Fig. 12A, C, and E, we plot the position of LBCVs in 3dimensional space relative to the emitter, detector, and channel. For
easier visualization, in Fig. 12 B, D, and F, we rotate the voxels into the
vertical plane that is defined by the emitter, the detector, and the
channel projection onto the brain surface. In Fig. 12A and B, we plot
LBCVs for all 1176 instances; in Fig. 12C–F, we only plot LBCVs which
have a BOLD-oxy-Hb correlation above a specified threshold (0.2 for C,
D and 0.4 for E, F). We note that the voxels seem to be randomly
distributed in this region and do not form a clean path.
We summarized the spatial distribution of LBCVs relative to the
channels in Table 2. The average scalp–brain distance is 15 mm. The
average distance between the LBCV and the channel projection on
the brain surface is 12 mm. The LBCV is on average 5 mm from the
brain surface. However, each of the distances shows considerable
variance (Fig. S4). We also found that the scalp–brain distance affects
the spatial location of LBCVs (Fig. S5).
567
Determining the optimal ROI shape
While LBCV distribution affords a general idea of NIRS measurement location, in reality photons always travel through multiple
voxels. To find the optimal photon path we created ROIs with various
shapes and studied how fMRI signals within these regions correlated with NIRS signals. While there are an infinite number of
possible shapes, we chose to focus on three: sphere, spherical shell,
and elliptical ring.
For spherical ROIs (Fig. 13A), we found that the best correlation
occurs, on average, when the sphere has a radius of 5 voxels (Fig. S6B).
For spherical shell-shaped ROIs, the best correlation occurs when the
radius is 4 voxels (~14 mm, Fig. S6C). This indicates that, on average,
voxels along a path 4 voxels away from the center of the channel
projection contribute most to the correlation.
We also systematically evaluated the fMRI-oxy-Hb correlation
using elliptical rings (Fig. 13B). The correlation reaches a peak when
the depth is 4 voxels and changes little as depth increases (Fig. S7B).
The best angle is near 0 (i.e., in the plane of emitter and detector),
ranging from −30° to 30° (Fig. S7C).
Among the different ROI shapes we studied, which one correlates
most strongly with the NIRS signal? For each instance, we calculated
correlations between BOLD and oxy-Hb for the range of possible
dimensions of each of the shapes. For each instance/shape, we
identified the highest correlation, which we refer to as the ‘best
correlation.’ We then averaged these ‘best correlations’ over all
instances (Fig. 13C). Subsets of the data were thresholded based on
the correlation with the single-voxel ROI. We find that an elliptical
ring, on average, is a better shape than shells or spheres, but it is not
better for all instances.
585
GLM-based analyses in each modality
614
A question of interest to NIRS researchers is whether similar
conclusions would be drawn from the same population and task,
measured with NIRS or fMRI. To answer this question, we performed
standard general linear model analyses on both NIRS and fMRI data
(Fig. 14). fMRI images are masked to show only locations measured
with the NIRS patches (see unmasked activations in Fig. S3). We found
that for the finger tapping task, in which the active region is on the
615
616
563
564
565
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
1.5
BOLD CNR
BOLD-NIRS correlation
552
553
X. Cui et al. / NeuroImage xxx (2010) xxx–xxx
0.6
0.5
0.4
p < 10-10
0.3
p < 10-10
0.2
p = 2x10-3
0.1
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
0
-1
-0.5
0
0.5
1
1.5
deoxy-Hb CNR
Fig. 11. CNR affects fMRI–NIRS correlation. The fMRI–NIRS correlation is more likely to
be low when the task-related CNR is low. (A) BOLD-oxy-Hb correlation (red), BOLDdeoxy-Hb correlation (blue), and BOLD-total-Hb correlation (green) as a function of
BOLD CNR. (B) BOLD–NIRS correlation as a function of oxy-Hb CNR. (C) BOLD–NIRS
correlation as a function of deoxy-Hb CNR.
Please cite this article as: Cui, X., et al., A quantitative comparison of NIRS and fMRI across multiple cognitive tasks, NeuroImage (2010),
doi:10.1016/j.neuroimage.2010.10.069
617
618
619
620
621
X. Cui et al. / NeuroImage xxx (2010) xxx–xxx
A
B
0
-5
0
-10
-10
All
11
-15
-20
-20
-30
30
-25
-30
20
10
0
-10
-20
-20
-10
0
10
-35
20
-30
C
D
-20
-10
0
10
20
30
-20
-10
0
10
20
30
-20
-10
0
10
20
30
0
-5
0
-10
-10
r>0.2
-15
-20
-20
-30
30
-25
-30
20
10
0
-10
-20
-20
-10
0
10
20
30
-35
-30
E
F
0
-5
0
-10
-10
r>0.4
-15
-20
-20
-30
30
-25
-30
20
10
0
-10
-20
-30 -30
-20
-10
0
10
20
30
-35
-30
Unit: mm
Fig. 12. Distributions of LBCVs relative to the probes and channels; each of the 1176 instances is shown (see Table 1 for details on the number of instances). (A) The location of LBCVs
in 3-D space. (B) The voxels have been rotated into the plane defined by the probes and the brain projection point. (C and D) The subset of voxels with (BOLD-oxy-Hb) r N 0.2. (E and
F) The subset of voxels with (BOLD-oxy-Hb) r N 0.4. Red circles indicate the emitter, blue indicate the detector, and black indicate the channel. LBCVs were determined based on
BOLD-oxy-Hb correlations; darker pink indicates a higher correlation value. (For interpretation of the references to colour in this figure legend, the reader is referred to the web
version of this article.)
622
623
624
625
626
t2:1
brain surface, the activation pattern in motor cortex is similar across
modalities. For cognitive tasks, the pattern of significant activations is
qualitatively similar in some parietal and frontal regions, suggesting
that similar conclusions may be drawn, but with inferior resolution in
the NIRS data.
Table 2
The mean value of the distances and angle of the distribution of the LBCVs. Please refer
to Fig. 2 for the meaning of symbols. Unit for distance is mm, and unit for angle is
degree.
t2:2
t2:3
t2:4
t2:5
t2:6
t2:7
t2:8
t2:9
t2:10
t2:11
t2:12
Oxy
All
N0.4
N0.5
N0.6
Deoxy
All
N0.4
N0.5
Sample size
a
b
c
d
e
f
α
1160
133
29
4
16
15
14
16
18
12
11
11
7
5
5
6
16
10
10
7
27
20
20
20
31
24
23
25
129
135
132
146
1080
125
39
16
14
14
21
11
11
6
5
5
19
9
9
31
19
19
33
22
23
127
132
135
Discussion
627
In this study we performed a comprehensive comparison of
hemodynamic signals measured with NIRS and fMRI in both temporal
and spatial domains, across multiple brain locations, and with a battery
of cognitive tasks. We found that, while many channels showed strong
correlations between BOLD and oxy- and deoxy-Hb, there was a wide
range in correlation values. We identified several factors which
contributed to variability of correlations, including scalp–brain
distance and CNR. We found that BOLD–NIRS correlations did not
depend significantly on the cognitive task. In the spatial domain, we
conducted an empirical exploration of the spatial correspondence of
NIRS and fMRI and found that, among the shapes we tested, a path
traversing an elliptical, annular ring between the emitter and detector
generated a time course that correlated most strongly with fMRI-BOLD.
We note, however, that this spatial correlation is dependent on the
specific probe geometry and placement used in our study.
Previous studies using simultaneous acquisition have mainly
focused on motor (Okamoto et al., 2004; Kleinschmidt et al., 1996;
Mehagnoul-Schipper et al., 2002; Toronov et al., 2001a, 2003;
628
Please cite this article as: Cui, X., et al., A quantitative comparison of NIRS and fMRI across multiple cognitive tasks, NeuroImage (2010),
doi:10.1016/j.neuroimage.2010.10.069
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
12
X. Cui et al. / NeuroImage xxx (2010) xxx–xxx
A
C
B
emitter
detector
Mean best correlation
0.7
0.6
r>0.6
r>0.5
r>0.4
r>0.3
r>0.2
r>0.1
r>0.0
0.5
0.4
depth
0.3
single
sphere ellipse
sphere
voxel
shell
ring
Fig. 13. BOLD–NIRS correlation as a function of ROI shape. (A) Schematic visualization
through a section of the spherical shell ROI. The channel marker (point A) is projected
to the brain surface (point B). Spherical or shell-shaped ROIs are grown from point
B with different radii; voxels outside the brain are discarded. (B) Schematic
visualization of the elliptical ROI. The locations of the emitter (red) and detector
(blue) are projected to the brain surface. With the two projection points as antipodal
points, elliptical ring-shaped ROIs are generated with 10 different depths (1–10 voxels)
and 7 different angles off the vertical plane (−90° to 90° with step size 30°).
(C) Comparison of BOLD-oxy-Hb correlations for different shapes. The single-voxel ROI
(LBCV) is also plotted for reference. Each color indicates a subset of correlations, with
cutoff threshold based on the LBCV correlation. For example, r N 0.3 means that we
selected all instances whose LBCV correlation is greater than 0.3 and calculated the
mean maximum correlation across all shapes. Circle areas are proportional to sample
size. (For interpretation of the references to colour in this figure legend, the reader is
referred to the web version of this article.)
646
647
648
649
650
651
Strangman et al., 2002; Boas et al., 2003; Hoge et al., 2005) and visual
(Toronov et al., 2007; Zhang et al., 2005a, 2006) paradigms and
regions. A strength of the present study is the breadth of both
cognitive tasks and brain regions simultaneously recorded with fMRI
and NIRS. In particular, we recorded from both lateral frontal and
parietal regions with tasks requiring response inhibition, working
memory, and visuospatial skills. As a result, we are able to generalize
our findings to tasks that typically result in smaller effect sizes than do
motor tasks. This is increasingly important as NIRS comes into use for
brain mapping studies across a wide range of cognitive domains.
Recording during multiple tasks and from multiple locations also
meant that we had a large number of instances, or pairs of fMRI/NIRS
time courses, to compare. This enabled us to perform higher-powered
statistical comparisons than had previously been possible.
NIRS is unique among imaging modalities in that it measures
changes in both oxy- and deoxy-Hb concentrations. Kleinschmidt
et al. (1996) found that increased BOLD response in motor cortex was
accompanied by decreased deoxy-Hb during a motor task. Subsequent
studies have found strong correlations between oxy-Hb and fMRI
BOLD. Strangman et al. (Strangman et al., 2002) found highly variable
correlations with deoxy-Hb and, after accounting for several possible
sources of error, found that the strongest correlations were with oxyHb. The authors suggest that this may be due to slightly higher signalto-noise in the oxy-Hb measure. In our data, we find that oxy-Hb has a
significant but small advantage over deoxy-Hb in terms of correlation
with BOLD, but no significant difference in CNR. A lack of clear
difference is perhaps not surprising given that oxy- and deoxy-Hb are
highly negatively correlated (Cui et al., 2010).
We found that task-related CNR was one factor which influenced
the strength of BOLD–NIRS correlations. Nonetheless, regions showing low task-related CNR also occasionally showed strong correlations. In non-task responsive regions, physiological (cardiac and
respiratory functions (Chang et al., 2009)) or thermal processes likely
contribute a higher proportion of signal variance. Concurrent
measurement of physiological responses could help to reduce such
errors (Tachtsidis et al., 2009; Glover et al., 2000); also see Strangman
et al., 2003) for a discussion of additional sources of error in NIRS
signal.
NIRS measurements rely on the relatively low absorption of
biological tissue to infrared light and the large difference in the
absorption spectra of oxy-Hb and deoxy-Hb in this range. As photons
are emitted they enter the head and traverse several layers: skin, skull,
and CSF, before penetrating the surface of the brain. Previous empirical
and simulation studies have addressed the issue of spatial resolution
of NIRS and the most likely shape for the photon path (Okada et al.,
Fig. 14. Activation maps measured with fMRI and NIRS. The channel positions projected onto the brain surface were normalized to a standard brain. For NIRS measurements, cubic
spline interpolation was applied to estimate the contrast value in voxels between channels. fMRI activations were masked to show only regions covered by the NIRS probes; unmasked
activation patterns are shown in Fig. S3. tap, tapping of fingers on left hand; nog, go/no-go; jlo, judgment of line orientation; vis, visuospatial N-back working memory task.
Please cite this article as: Cui, X., et al., A quantitative comparison of NIRS and fMRI across multiple cognitive tasks, NeuroImage (2010),
doi:10.1016/j.neuroimage.2010.10.069
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
X. Cui et al. / NeuroImage xxx (2010) xxx–xxx
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
1995, 1997; Okada and Delpy, 2003) and estimated the likely
contribution of specific brain regions to the NIRS signal (Zhang et al.,
2005a). Here, we have adopted a different approach, directly
correlating fMRI signals for different ROI shapes with NIRS measurements, in order to perform a quantitative comparison of these two
methodologies. With this approach we were able to identify the most
likely of several possible path shapes.
One particularly salient finding was that increased scalp–brain
distance reduces NIRS CNR and correlation with BOLD and seems to
affect NIRS–fMRI correlation more strongly than the experimental
paradigm. Variability in scalp–brain distance across individuals may
be one factor contributing to variance in signal quality from different
individuals. While head shapes will undoubtedly vary, our analysis of
the standard template brain indicated that frontal regions have
shorter scalp–brain distance than medial parietal regions. This has
important implications for the choice of paradigm and recording sites
for NIRS studies. Frontal regions that can be recorded by placing NIRS
probes on the forehead also have the advantage of being hairless,
which allows increased contact of the probes with the scalp. We
caution that this does not mean that one cannot obtain reliable
measurements from parietal regions, as previous studies have done so
(Herrmann et al., 2005), but rather that the likelihood of weakened
CNR in some regions should inform the task design and interpretation
of results. One may also consider the possibility of altering the
emitter–detector distance when recording in certain regions in order
to deepen the photon path. We also note that in our data we found
qualitatively better correspondence between BOLD and NIRS signals
in parietal regions for some tasks, perhaps indicating that despite
larger scalp–brain distance, the activated tissue (i.e., with higher
BOLD CNR) was closer to the brain surface than in frontal regions.
All of the tasks in the present study were set up as block designs in
order to maximize the signal-to-noise ratio. While this was entirely
appropriate for a preliminary investigation of fMRI/NIRS correlations,
it would also be interesting to compare responses during an eventrelated design. Given the reduced CNR of NIRS relative to fMRI, it is of
interest whether event-related designs are practical. Several studies
have successfully reported results from event-related NIRS paradigms
(Nishimura et al., 2010; Boecker et al., 2007; Lee et al., 2008). Plichta
et al. (Plichta et al., 2006) found strong reproducibility of group-level
event-related activations of visual cortex, but poor reproducibility at
the single-subject level. A direct comparison of event-related NIRS
and fMRI could be informative on this point.
We note that our spatial analysis does not provide a definite
description of the photon path, but merely a sense of the spatial
correspondence with fMRI. We also note that we did not incorporate
information about underlying sulcal geometry into our estimation of
the NIRS photon path. Previous studies have used segmented MRI
images, combined with Monte Carlo simulations of light transport, to
estimate the spatial distribution of the source of the NIRS signal in
three dimensions (Zhang et al., 2005a; Toronov et al., 2007). In this
study we favored a purely empirical approach, correlating the signal
along different paths with the NIRS response. However, a more
informed model of the underlying geometry could in principle help
constrain the photon path and potentially provide information about
the strength of NIRS signals from the surfaces of gyri relative to sulci.
Finally, we analyzed the NIRS and fMRI data separately to establish
whether similar conclusions would be drawn from data collected on
the same participants and using the same tasks. We found
qualitatively similar activations in the NIRS and fMRI analyses, though
with higher significance in the fMRI data. This analysis emphasizes the
utility of a structural MRI in conjunction with NIRS data, as the
activations are much more difficult to localize precisely without this
information. We also note that in some instances (e.g., tapping task),
NIRS signals show significant activation in regions that do not show
significant activations in the fMRI data (e.g., frontal regions). This
indicates that, due to lower CNR, NIRS data may be more susceptible
13
to false positives. Correcting for physiological noise (Tachtsidis et al.,
2009; Glover et al., 2000) or filtering out global signals using principal
components analysis (PCA) (Virtanen et al., 2009) could potentially
reduce the occurrence of false positives. However, as the purpose of
the present study was a comparison of NIRS and fMRI signals, we
chose to apply similar band-pass filtering to data collected from both
modalities to correct for physiological noise.
Simultaneous acquisition of NIRS and fMRI has previously been
used to elucidate the biophysics of the BOLD response (Hoge et al.,
2005; Toronov et al., 2003; Kleinschmidt et al., 1996), to enhance fMRI
measurements with information about blood flow (MehagnoulSchipper et al., 2002), or to establish correlations between fMRI and
NIRS signals in terms of spatial location or amplitude (Toronov et al.,
2001a, 2007; Strangman et al., 2002). A synthesis of the results of
these studies indicates that though the SNR of NIRS measurements
is lower, there is nonetheless a strong correlation between signals
measured with NIRS and fMRI. The present study makes several novel
contributions to this literature. First, we demonstrated that this
correlation is present during cognitive tasks that generate more
subtle activation changes relative to motor or visual tasks. We also
established that reduced CNR leads to reduced correlation between
NIRS and fMRI signals. We investigated the relationship between head
geometry and NIRS–fMRI correlations and found that scalp–brain
distance was an important factor in determining the strength of this
correlation. We also found that the NIRS signal was most strongly
correlated with the fMRI signal along an elliptical path. Since NIRS has
become increasingly popular for brain mapping studies across a range
of cognitive domains, studies such as this should help to build
confidence in NIRS as a complement to fMRI.
Supplementary data to this article can be found online at
doi:10.1016/j.neuroimage.2010.10.069.
757
758
Acknowledgments
788
This work is supported by S10 RR024657 (ALR PI), P41 RR009874
(GHG), the Stanford Institute for Neuro-Innovation and Translational
Neurosciences (SINTN) fellowship (X.C. and A.L.R.), NARSAD Young
Investigator's Award, and Stanford University Lucas Center Radiology
Seed Grant. We thank Dr. Fumiko Hoeft for her invaluable assistance
in planning the experiments and solving technical issues. We thank
Sebastian Heinzel for valuable suggestions and comments on this
project. We thank Joerg Schnackenberg and Hitachi Medical Inc. for
invaluable assistance with hardware setup for this study.
789
References
798
Abdelnour, F., Schmidt, B., Huppert, T.J., 2009. Topographic localization of brain
activation in diffuse optical imaging using spherical wavelets. Phys. Med. Biol. 54,
6383–6413.
Atsumori, H., Kiguchi, M., Obata, A., Sato, H., Katura, T., Funane, T., Maki, A., 2009.
Development of wearable optical topography system for mapping the prefrontal
cortex activation. Rev. Sci. Instrum. 80, 043704.
Boas, D.A., Strangman, G., Culver, J.P., Hoge, R.D., Jasdzewski, G., Poldrack, R.A., Rosen, B.
R., Mandeville, J.B., 2003. Can the cerebral metabolic rate of oxygen be estimated
with near-infrared spectroscopy? Phys. Med. Biol. 48, 2405–2418.
Boecker, M., Buecheler, M.M., Schroeter, M.L., Gauggel, S., 2007. Prefrontal brain
activation during stop-signal response inhibition: an event-related functional nearinfrared spectroscopy study. Behav. Brain Res. 176, 259–266.
Buxton, R.B., Wong, E.C., Frank, L.R., 1998. Dynamics of blood flow and oxygenation
changes during brain activation: the balloon model. Magn. Reson. Med. 39,
855–864.
Chang, C., Cunningham, J.P., Glover, G.H., 2009. Influence of heart rate on the BOLD
signal: the cardiac response function. Neuroimage 44, 857–869.
Cope, M., Delpy, D.T., 1988. System for long-term measurement of cerebral blood and
tissue oxygenation on newborn infants by near infra-red transillumination. Med.
Biol. Eng. Comput. 26, 289–294.
Cope, M., 1991. The application of near infrared spectroscopy to non-invasive
monitoring of cerebral oxygenation in the newborn infant. University College
London.
Coyle, S., Ward, T., Markham, C., McDarby, G., 2004. On the suitability of near-infrared
(NIR) systems for next-generation brain–computer interfaces. Physiol. Meas. 25,
815–822.
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
Please cite this article as: Cui, X., et al., A quantitative comparison of NIRS and fMRI across multiple cognitive tasks, NeuroImage (2010),
doi:10.1016/j.neuroimage.2010.10.069
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
790
791
792
793
794
795
796
797
14
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
X. Cui et al. / NeuroImage xxx (2010) xxx–xxx
Coyle, S.M., Ward, T.E., Markham, C.M., 2007. Brain–computer interface using a
simplified functional near-infrared spectroscopy system. J. Neural Eng. 4, 219–226.
Cui, X., Bray, S., Reiss, A.L., 2010. Functional near infrared spectroscopy (NIRS) signal
improvement based on negative correlation between oxygenated and deoxygenated hemoglobin dynamics. Neuroimage 49, 3039–3046.
Emir, U.E., Ozturk, C., Akin, A., 2008. Multimodal investigation of fMRI and fNIRS derived
breath hold BOLD signals with an expanded balloon model. Physiol. Meas. 29, 49–63.
Glover, G.H., Li, T.Q., Ress, D., 2000. Image-based method for retrospective correction of
physiological motion effects in fMRI: RETROICOR. Magn. Reson. Med. 44, 162–167.
Glover, G.H., Law, C.S., 2001. Spiral-in/out BOLD fMRI for increased SNR and reduced
susceptibility artifacts. Magn. Reson. Med. 46, 515–522.
Haberecht, M.F., Menon, V., Warsofsky, I.S., White, C.D., Dyer-Friedman, J., Glover, G.H.,
Neely, E.K., Reiss, A.L., 2001. Functional neuroanatomy of visuo-spatial working
memory in Turner syndrome. Hum. Brain Mapp. 14, 96–107.
Herrmann, M.J., Ehlis, A.C., Wagener, A., Jacob, C.P., Fallgatter, A.J., 2005. Near-infrared
optical topography to assess activation of the parietal cortex during a visuo-spatial
task. Neuropsychologia 43, 1713–1720.
Hoeft, F., Hernandez, A., Parthasarathy, S., Watson, C.L., Hall, S.S., Reiss, A.L., 2007.
Fronto-striatal dysfunction and potential compensatory mechanisms in male
adolescents with fragile X syndrome. Hum. Brain Mapp. 28, 543–554.
Hoge, R.D., Franceschini, M.A., Covolan, R.J.M., Huppert, T., Mandeville, J.B., Boas, D.A.,
2005. Simultaneous recording of task-induced changes in blood oxygenation,
volume, and flow using diffuse optical imaging and arterial spin-labeling MRI.
Neuroimage 25, 701–707.
Honda, Y., Nakato, E., Otsuka, Y., Kanazawa, S., Kojima, S., Yamaguchi, M.K., Kakigi, R.,
2010. How do infants perceive scrambled face? A near-infrared spectroscopic
study. Brain Res. 1308, 137–146.
Huppert, T.J., Allen, M.S., Benav, H., Jones, P.B., Boas, D.A., 2007. A multicompartment
vascular model for inferring baseline and functional changes in cerebral oxygen
metabolism and arterial dilation. J. Cereb. Blood Flow Metab. 27, 1262–1279.
Huppert, T.J., Allen, M.S., Diamond, S.G., Boas, D.A., 2009. Estimating cerebral oxygen
metabolism from fMRI with a dynamic multicompartment Windkessel model.
Hum. Brain Mapp. 30, 1548–1567.
Huppert, T.J., Diamond, S.G., Boas, D.A., 2008. Direct estimation of evoked hemoglobin
changes by multimodality fusion imaging. J. Biomed. Opt. 13, 054031.
Huppert, T.J., Hoge, R.D., Dale, A.M., Franceschini, M.A., Boas, D.A., 2006a. Quantitative
spatial comparison of diffuse optical imaging with blood oxygen level-dependent
and arterial spin labeling-based functional magnetic resonance imaging. J. Biomed.
Opt. 11, 064018.
Huppert, T.J., Hoge, R.D., Diamond, S.G., Franceschini, M.A., Boas, D.A., 2006b. A
temporal comparison of BOLD, ASL, and NIRS hemodynamic responses to motor
stimuli in adult humans. Neuroimage 29, 368–382.
Kennan, R.P., Kim, D., Maki, A., Koizumi, H., Constable, R.T., 2002. Non-invasive
assessment of language lateralization by Transcranial near infrared optical
topography and functional MRI. Hum. Brain Mapp. 16, 183–189.
Kesler, S.R., Haberecht, M.F., Menon, V., Warsofsky, I.S., Dyer-Friedman, J., Neely, E.K.,
Reiss, A.L., 2004. Functional neuroanatomy of spatial orientation processing in
Turner syndrome. Cereb. Cortex 14, 174–180.
Kleinschmidt, A., Obrig, H., Requardt, M., Merboldt, K.D., Dirnagl, U., Villringer, A.,
Frahm, J., 1996. Simultaneous recording of cerebral blood oxygenation changes
during human brain activation by magnetic resonance imaging and near-infrared
spectroscopy. J. Cereb. Blood Flow Metab. 16, 817–826.
Lee, J., Folley, B.S., Gore, J., Park, S., 2008. Origins of spatial working memory deficits in
schizophrenia: an event-related FMRI and near-infrared spectroscopy study. PLoS
ONE 3, e1760.
Lu, C.M., Zhang, Y.J., Biswal, B.B., Zang, Y.F., Peng, D.L., Zhu, C.Z., 2010. Use of fNIRS to
assess resting state functional connectivity. J. Neurosci. Meth. 186, 242–249.
Malonek, D., Dirnagl, U., Lindauer, U., Yamada, K., Kanno, I., Grinvald, A., 1997. Vascular
imprints of neuronal activity: relationships between the dynamics of cortical blood
flow, oxygenation, and volume changes following sensory stimulation. Proc. Natl
Acad. Sci. USA 94, 14826–14831.
Mehagnoul-Schipper, D.J., Colier, W.N.J.M., van Erning, L.J.T.O., Thijssen, H.O.M.,
Oeseburg, B., Hoefnagels, W.H.L., Jansen, R.W.M.M., 2002. Simultaneous measurements of cerebral oxygenation changes during brain activation by near-infrared
spectroscopy and functional magnetic resonance imaging in healthy young and
elderly subjects. Hum. Brain Mapp. 16, 14–23.
Nishimura, Y., Sugisaki, K., Hattori, N., Inokuchi, Y., Komachi, M., Nishimura, Y., Ogawa,
M., Okada, M., Okazaki, Y., Taki, W., et al., 2010. An event-related fNIRS
investigation of Japanese word order. Exp. Brain Res. 202, 239–246.
Obata, T., Liu, T.T., Miller, K.L., Luh, W.M., Wong, E.C., Frank, L.R., Buxton, R.B., 2004.
Discrepancies between BOLD and flow dynamics in primary and supplementary
motor areas: application of the balloon model to the interpretation of BOLD
transients. Neuroimage 21, 144–153.
Ogawa, S., Lee, T.M., Kay, A.R., Tank, D.W., 1990. Brain magnetic resonance imaging with
contrast dependent on blood oxygenation. Proc. Natl Acad. Sci. USA 87, 9868–9872.
Okada, E., Firbank, M., Schweiger, M., Arridge, S.R., Cope, M., Delpy, D.T., 1997.
Theoretical and experimental investigation of near-infrared light propagation in a
model of the adult head. Appl. Opt. 36, 21–31.
Okada, E., Firbank, M., Delpy, D.T., 1995. The effect of overlying tissue on the
spatial sensitivity profile of near-infrared spectroscopy. Phys. Med. Biol. 40,
2093–2108.
Okada, E., Delpy, D.T., 2003. Near-infrared light propagation in an adult head model. I.
Modeling of low-level scattering in the cerebrospinal fluid layer. Appl. Opt.
Okamoto, M., Dan, H., Shimizu, K., Takeo, K., Amita, T., Oda, I., Konishi, I., Sakamoto, K.,
Isobe, S., Suzuki, T., et al., 2004. Multimodal assessment of cortical activation during
apple peeling by NIRS and fMRI. Neuroimage 21, 1275–1288.
Plichta, M.M., Herrmann, M.J., Baehne, C.G., Ehlis, A.C., Richter, M.M., Pauli, P., Fallgatter,
A.J., 2006. Event-related functional near-infrared spectroscopy (fNIRS): are the
measurements reliable? Neuroimage 31, 116–124.
Power, S.D., Falk, T.H., Chau, T., 2010. Classification of prefrontal activity due to mental
arithmetic and music imagery using hidden Markov models and frequency domain
near-infrared spectroscopy. J. Neural Eng. 7, 026002.
Schroeter, M.L., Kupka, T., Mildner, T., Uludag, K., von Cramon, D.Y., 2006. Investigating
the post-stimulus undershoot of the BOLD signal–a simultaneous fMRI and fNIRS
study. Neuroimage 30, 349–358.
Sitaram, R., Zhang, H.H., Guan, C.T., Thulasidas, M., Hoshi, Y., Ishikawa, A., Shimizu, K.,
Birbaumer, N., 2007. Temporal classification of multichannel near-infrared
spectroscopy signals of motor imagery for developing a brain–computer interface.
Neuroimage 34, 1416–1427.
Steinbrink, J., Villringer, A., Kempf, F., Haux, D., Boden, S., Obrig, H., 2006. Illuminating
the BOLD signal: combined fMRI-fNIRS studies. Magn. Reson. Imaging 24,
495–505.
Strangman, G., Culver, J.P., Thompson, J.H., Boas, D.A., 2002. A quantitative comparison
of simultaneous BOLD fMRI and NIRS recordings during functional brain activation.
Neuroimage 17, 719–731.
Strangman, G., Franceschini, M.A., Boas, D.A., 2003. Factors affecting the accuracy of
near-infrared spectroscopy concentration calculations for focal changes in
oxygenation parameters. Neuroimage 18, 865–879.
Suda, M., Takei, Y., Aoyama, Y., Narita, K., Sato, T., Fukuda, M., Mikuni, M., 2010.
Frontopolar activation during face-to-face conversation: an in situ study using
near-infrared spectroscopy. Neuropsychologia 48, 441–447.
Tachtsidis, I., Leung, T.S., Chopra, A., Koh, P.H., Reid, C.B., Elwell, C.E., 2009. False
positives in functional near-infrared topography. Oxygen Transport To Tissue Xxx
Book Series. Adv. Exp. Med. Biol. 645, 307–314.
Tomioka, H., Yamagata, B., Takahashi, T., Yano, M., Isomura, A.J., Kobayashi, H., Mimura,
M., 2009. Detection of hypofrontality in drivers with Alzheimer's disease by nearinfrared spectroscopy. Neurosci. Lett. 451, 252–256.
Toronov, V., Walker, S., Gupta, R., Choi, J.H., Gratton, E., Hueber, D., Webb, A., 2003. The
roles of changes in deoxyhemoglobin concentration and regional cerebral blood
volume in the fMRI BOLD signal. Neuroimage 19, 1521–1531.
Toronov, V., Webb, A., Choi, J.H., Wolf, M., Michalos, A., Gratton, E., Hueber, D., 2001a.
Investigation of human brain hemodynamics by simultaneous near-infrared
spectroscopy and functional magnetic resonance imaging. Med. Phys. 28,
521–527.
Toronov, V., Webb, A., Choi, J.H., Wolf, M., Safonova, L., Wolf, U., Gratton, E., 2001b.
Study of local cerebral hemodynamics by frequency-domain near-infrared
spectroscopy and correlation with simultaneously acquired functional magnetic
resonance imaging. Opt. Express 9, 417–427.
Toronov, V.Y., Zhang, X., Webb, A.G., 2007. A spatial and temporal comparison of
hemodynamic signals measured using optical and functional magnetic resonance
imaging during activation in the human primary visual cortex. Neuroimage 34,
1136–1148.
Utsugi, K., Obata, A., Sato, H., Aoki, R., Maki, A., Koizumi, H., Sagara, K., Kawamichi, H.,
Atsumori, H., Katura, T., 2008. GO-STOP control using optical brain–computer
interface during calculation task. IEICE Trans. Commun. E91-B, 2133–2141.
Virtanen, J., Noponen, T., Merilainen, P., 2009. Comparison of principal and independent
component analysis in removing extracerebral interference from near-infrared
spectroscopy signals. J. Biomed. Opt. 14, 054032.
White, B.R., Snyder, A.Z., Cohen, A.L., Petersen, S.E., Raichle, M.E., Schlaggar, B.L., Culver,
J.P., 2009. Resting-state functional connectivity in the human brain revealed with
diffuse optical tomography. Neuroimage 47, 148–156.
Zhang, H., Zhang, Y.J., Lu, C.M., Ma, S.Y., Zang, Y.F., Zhu, C.Z., 2010. Functional
connectivity as revealed by independent component analysis of resting-state fNIRS
measurements. Neuroimage 51, 1150–1161.
Zhang, X.F., Toronov, V.Y., Webb, A.G., 2006. Integrated measurement system for
simultaneous functional magnetic resonance imaging and diffuse optical tomography in human brain mapping. Rev. Sci. Instrum. 77, 114301.
Zhang, X.F., Toronov, V.Y., Webb, A.G., 2005a. Simultaneous integrated diffuse optical
tomography and functional magnetic resonance imaging of the human brain. Opt.
Express 13, 5513–5521.
Zhang, Y., Brooks, D.H., Franceschini, M.A., Boas, D.A., 2005b. Eigenvector-based spatial
filtering for reduction of physiological interference in diffuse optical imaging. J.
Biomed. Opt. 10, 11014.
van der Zee, P., Arridge, S.R., Cope, M., Delpy, D.T., 1990. The effect of optode positioning
on optical pathlength in near infrared spectroscopy of brain. Adv. Exp. Med. Biol.
277, 79–84.
981
Please cite this article as: Cui, X., et al., A quantitative comparison of NIRS and fMRI across multiple cognitive tasks, NeuroImage (2010),
doi:10.1016/j.neuroimage.2010.10.069
903
904
905
906
907 Q2
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980